investment banking and analyst objectivity:
Post on 25-Jan-2015
393 Views
Preview:
DESCRIPTION
TRANSCRIPT
Investment Banking and Analyst Objectivity: Evidence on Forecasts and Recommendations of Analysts
Affiliated with M&A Advisors
By Adam C. Kolasinski+ and S.P. Kothari++
+Sloan School of Management 50 Memorial Drive, E52-458 Cambridge, MA 02459-1261
(617) 253-3919 ack@mit.edu
++Sloan School of Management 50 Memorial Drive, E52-325c Cambridge, MA 02459-1261
(617) 253-3919 kothari@mit.edu
October 2003
Previous research finds some evidence that analysts affiliated with equity underwriters issue more optimistic recommendations of client stock and forecasts of client earnings growth than those issued by analysts without such affiliations. Unfortunately, these studies are unable to determine whether such optimism is due to a conflict of interest problem, whereby analysts compromise their objectivity at the behest of underwriting clients, or selection bias, whereby investment banks employing analysts more bullish on a particular stock are more likely to underwrite it. Examining the issue of analyst bias in the M&A context allows us to shed some light on which explanation is correct.
1. Introduction
Analysts play an important role in the securities underwriting business, and this
role has become a topic of increasing interest to regulators and academics. Several
studies find evidence that analysts affiliated with investment banking firms (“affiliated
analysts”) issue positively biased recommendations and overly optimistic long-term
earnings growth forecasts of stocks underwritten by their employers.1 Lin, McNichols
and O’Brien (2003) find that affiliated analysts are slower to downgrade their
recommendations of client firms than unaffiliated analysts, and Bradshaw, Richardson
and Sloan (2003) find evidence that consensus analyst coverage of firms issuing
securities is more optimistic than of firms not issuing any securities. Consistent with the
academic research suggesting that economic incentives stemming from investment-
banking relations and brokerage commission revenues optimistically skew the tone of
affiliated analysts’ research, the New York Attorney General recently reached a
settlement with several investment banks. As per the settlement, in addition to paying a
fine, the banks agreed to remove as a factor in analyst compensation the generation of
underwriting revenue. The investment banks settled, it appears, largely because of the
discovery of internal memos in which, in order to maintain their employers’ investment
banking relationships with the issuers, the analysts admitted to recommending stocks they
believed to be unsound investments.2
In the equity underwriting context, corporate managers may seek optimistic
analyst coverage in the hope that it would enable the company to issue shares at a higher
1 See Lin and McNichols (1998), Michaely and Womack (1999), Dechow, Hutton and Sloan (2000), Dugar and Nathan (1995). 2 “Chronology of the Merrill Lynch Probe.” The Associated Press. May 21, 2002.
2
price. Hence, it is alleged that managers reward optimistic analyst coverage by giving
their equity underwriting business to the investment-banking firm employing the analyst.
Seeking to gain such rewards, analysts might compromise their objectivity when issuing
forecasts and recommendations of firms from which their employers’ investment banking
departments are seeking to obtain business. Optimistic analyst coverage, however,
benefits managers in ways in addition to pushing up new issue prices. Since previous
research demonstrates that favorable analyst coverage is associated with superior stock
performance (e.g., Womack, 1996), managers have an incentive to do whatever they can
to generate favorable analyst coverage, including rewarding firms who provide such
coverage with M&A business.
In this paper we examine whether M&A relationships affect analyst objectivity.
Studying analyst objectivity in the M&A context has certain advantages. First, the M&A
context allows for more powerful tests of the effect of investment banking revenue on
analyst forecasts and recommendations than does the equity issuance context. The tests
are more powerful because depending on the M&A relationship with the target or
acquirer the analyst might have an incentive to bias the research optimistically or
pessimistically. Second, M&A transactions are much more frequent than equity
issuances, and M&A fees make up at least as large a portion of investment banking
revenue as underwriting fees. Thus, the incentives M&A fees provide for analysts to
compromise their objectivity are potentially as large as that of stock issuances. Third, the
number of firms engaging in M&A transactions is far greater than the number issuing
equity. Therefore, M&A-driven analyst optimism or pessimism, if it were to exist, would
be a far more pervasive problem than equity underwriting-driven analyst optimism.
3
Finally, and most importantly, unlike the equity underwriting context, the M&A context
allows us to discriminate between competing hypotheses that predict a positive
association between investment banking relationships and analyst optimism or
pessimism.
There are two possible hypotheses consistent with the finding that affiliated
analysts are more optimistic in their coverage of underwriting clients. The first
hypothesis, henceforth the “conflict of interest hypothesis,” postulates that such analysts
deliberately bias their reports in order to curry favor with the clients (or potential clients).
The second hypothesis, henceforth the “selection bias hypothesis,” postulates that
analysts are objective, but if an analyst is optimistic about the prospects of a given firm,
the analyst’s employer is more likely to get the investment banking business of that firm.
Selection bias may result from investment banks’ unwillingness to underwrite stocks if
their analysts are not optimistic, or it may be result from CFOs’ unwillingness to hire
investment banks with pessimistic analysts. Regardless, the M&A context allows to
discriminate between the conflict of interest and selection bias hypotheses because, for
reasons explained in the last part of Section 2, the selection bias hypothesis would be an
unlikely explanation of affiliated analyst optimism (or pessimism), were to exist, in the
M&A context. Furthermore, we show in Section 2 that there are situations within the
M&A context in which selection bias cannot possibly explain hypothetical analyst
optimism (or pessimism).
In order to test the conflict of interest hypotheses and attempt to distinguish its
effects from the selection bias hypothesis, we obtain a sample of every M&A deal
completed between 1993 and 2001. Using OLS analysis, we test whether analysts
4
working for firms which collect M&A fees or have an M&A relationships issue more or
less optimistic forecasts relative to consensus. Notwithstanding high power and large
sample size, we fail to find a significant association. Using ordered logit analysis, we
conduct a similar test on the recommendations of target or acquirer stock issued by
affiliated analysts and find similar results. As will be explained in the next section, there
is no reason to believe that there is any less potential for conflict of interest in the M&A
context than in the equity underwriting context. Therefore, these results indicate that the
association between analyst optimism and underwriting relationships found in previous
studies is due mainly to selection bias rather than conflict of interest. We must stress that
our research does not rule out the existence of conflict of interest. It is possible, given
our results, that a conflict of interest exists, but regulations already in place prevent it
from tainting analyst forecasts and recommendations.
Section 2 explains in detail the advantages of studying analyst objectivity in the
M&A context and describes how the case of M&A relationships allows us to distinguish
between the selection bias and conflict of interest explanations of the relative optimism of
analysts affiliated with investment banks. Section 3 describes our data and presents
descriptive statistics and preliminary analysis based on such statistics. Section 4 describes
our ordinary least squares analysis of EPS and growth forecasts and presents the results.
Section 5 describes our ordered logit analysis of recommendations and presents the
results. Section 6 concludes.
2. M&A Context to Study Analyst Objectivity: Development of Competing Hypotheses
5
In this section we discuss in detail four topics to which we alluded above. First,
we discuss how the M&A context allows us to design more powerful tests. Second, we
explain how the potential for conflict of interest is just as strong, if not stronger in the
M&A context than in the equity underwriting context. Third, we demonstrate how much
more pervasive M&A advisory activity is relative to equity underwriting. Finally, we
show that the M&A context allows us to distinguish between competing hypotheses.
2.1 More Powerful Tests
Previous research suggests that analysts optimistically bias their research to
generate equity underwriting business (eg. Michaely and Womack, 1999). There are
reasons specific to the M&A context that would cause managers to want optimistic
coverage of their own firm and, in some cases, pessimistic coverage of the counterparty’s
firm. If optimistic (pessimistic) coverage tends to positively (negatively) influence the
stock price, favorable coverage of the acquirer around the time of a stock deal would tend
to make the terms of the deal better for the acquirer. Pessimistic coverage of the acquirer
before the transaction is complete would make the terms better for a target in a stock
deal, since target shareholders want to get as many acquirer shares as possible. Similarly,
optimistic coverage of the target firm would be good for target shareholders but bad for
acquiring shareholders in both cash and stock deals. Therefore, if analysts bias their
reports to please M&A clients, in some instances their reports will be optimistic, and in
other instances they will be pessimistic, depending on which advisor the analyst is
working for, whose stock he is covering, and the type of deal. This greater variety of
predicted biases allows us to design more powerful tests in the M&A context than in the
equity underwriting context.
6
2.2 The Potential for Conflict of Interest: a Comparison of the M&A and Equity Underwriting Contexts.
Stock issuances, which generate underwriting revenue, are a rare event in the life
of a firm; the vast majority of firms only have one, and very few have more than two. By
contrast many firms make multiple acquisitions in their lifetime, and at any given time
the probability of being acquired is high for a large number of firms. The volume of
M&A activity is an order of magnitude higher than equity underwriting. For instance, in
1999, one of the biggest equity issuance years in history, our analysis of the SDC equity
issuance database shows that public equity offerings in which an investment bank was
hired raised just under $200 billion in aggregate proceeds. By contrast, in 1999 the
aggregate transaction value of M&A deals in which at least one investment bank was
hired as an advisor exceeded $1.8 trillion.
Not only are M&A transactions far more pervasive than equity issuances, M&A
fees in aggregate are still just as important if not more important to investment banking
firms than are equity underwriting fees. According to Freeman & Co. estimates, and as
illustrated in figure 1, in every year since 1994, M&A fees in the US have been at least as
large as equity underwriting fees, and in recent years significantly larger. Since M&A
fee revenues are just as important to the investment banking business as are underwriting
revenues it would be highly unlikely that managers would use the promise of equity
underwriting fees and not M&A fees to coax investment banks into providing favorable
desired analyst coverage. Since M&A transactions occur with greater fr equency, the
promise of M&A business would give managers of acquiring firms even more leverage
since such a promise would mean repeat business for the investment bank, rather than a
one-time equity underwriting fee. Furthermore, there is no reason why profit-
7
maximizing investment banks that pressure analysts to provide desired coverage in order
to obtain underwriting business would not do the same in order to obtain M&A business.3
As counter-argument, one might argue that analysts are involved in the equity
issuance process but are kept out of the M&A process. Hence analysts find out about
equity deals earlier than M&A deals, about which they only officially learn after the
public announcement. Hence analysts have a greater opportunity to compromise their
objectivity in the equity issuance context than in the M&A context. Conversations with
practitioners and personal experience in the investment banking industry on the part of
one of the authors, however, reveal that this argument has little validity.4 Analysts are
typically brought on board equity deals late in the process, just before the deals are made
public, and hence they do not learn about them much earlier than the public, at least
officially. Furthermore, SEC regulations prohibit analysts from reporting on client firms
from the time they have been brought on board an equity deal until the end of the SEC-
mandated “quiet period” 25 days after the deal.5 Hence there is no more opportunity for
analysts to bias their forecasts and recommendations in equity deals than in M&A deals.
In fact, there may be a greater opportunity to do so in the M&A context since there are no
regulations or internal policies prohibiting analysts affiliated with M&A advisors from
3 There may be institutional reasons why investment banks may pressure analysts to generate underwriting business but not M&A business. For instance, the M&A and research departments may not be linked in the same manner as the underwriting and research departments, so the structure of the institution may not allow for analysts to be rewarded for generating M&A revenue as for underwriting revenue, even though such rewards would be profit-maximizing. We ignore such questions here, leaving them for future research. 4 We are greatful to Kevin Rock for his elucidation of the institutional deatails. Adam Kolasinski worked in the investment banking division of Wasserstein Perrella & Co. from 1998-1999. 5 See SEC rule 174 of the Securities Act of 1933 and the 1988 revision of rule 174.
8
issuing reports on client or counterparty firms at any time.6 The New York State
Attorney General considers potential for conflict of interest problems generated by M&A
relationships strong enough that the state’s settlement with Merrill Lynch requires
analysts to disclose in their reports M&A rela tionships as well as equity underwriting
relationships.7
2.3 The Pervasiveness of M&A Activity
Analyst bias in the M&A context is even more important from the standpoint of
public policy. The proportion of firms undergoing or likely to undergo an equity issue at
any given point in time is small relative to the total number of public firms. Using the
CRSP and SDC databases, we calculate that in 1999, one of the biggest equity issuance
years in history, the year-end aggregate market capitalization of US firms undergoing at
least one equity offering was $2.25 trillion, which, while not an inconsiderable number,
was only 13% of the value of aggregate US equity market capitalization.8 By contrast,
the total 1999 year-end aggregate market value of public firms which engaged in an
M&A transaction in which an advisor was hired was $9.21 trillion, or 55% of aggregate
US market capitalization.9 Figure 2 plots the historical share of aggregate US equity
market capitalization represented by firms undergoing equity offerings and M&A
6 Investment banks almost universally have compliance policies that restrict analysts’ direct communications with M&A client firms, but lack of such communications do not stop them from issuing reports about M&A clients. 7 John Goff, “Wall? What Chinese Wall?” CFO.com, April 22, 2002. 8 Year-end market capitalizations obtained from CRSP. If a year-end market capitalization was not available, the market-capitalization on the last trading day of the year was used. 9 Year-end market capitalizations obtained from CRSP. If a year-end market capitalization was not available, the market-capitalization on the last trading day of the year was used.
9
transactions. If M&A relationships taint analyst forecasts and recommendations, there
would potentially exist a policy problem of massive proportions.
2.4 Distinguishing Between Competing Hypotheses
As discussed in the introduction, the association between analyst optimism and
underwriting relationships found in previous research is consistent with both the conflict
of interest hypothesis and the selection bias hypothesis. In this subsection, we note how
all previous research has failed to distinguish between these two hypotheses, discuss in
detail how each may or may not be valid in the M&A context, and demonstrate how, in
some circumstances, the two hypotheses make different predictions in the M&A context.
2.4.1 Inability of Previous Research to Distinguish Between Competing Hypotheses
To date, no research that compares the forecasts and recommendations of
affiliated analysts with those of unaffiliated analysts has managed to empirically
distinguish between these two hypotheses. With most papers the inability to distinguish
between hypotheses is obvious, and we shall not discuss them in detail here. There are,
however, two papers whose research design seems to distinguish between these
hypotheses. Below we explain why they do not.
Lin, McNichols, and O’Brien (2003), henceforth “LMO,” claim that their finding
that affiliated analysts are slower to downgrade client firms in response to bad news
constitutes evidence in favor of the conflict of interest hypothesis and cannot be
explained by the selection bias hypothesis. We disagree. An analyst who is optimistic
about a firm’s prospects is likely to give less weight to bad news, and hence would be
10
slower to downgrade a firm. Thus LMO’s cannot rule out selection bias as an
explanation for their results.
To justify their claim, LMO state, “To incorporate our evidence into the
underwriter selection story, one must assume that managers choose underwriters both on
their observable optimism and on the unobservable strength of their beliefs.” We believe
this argument to be flawed for two reasons. First, it is likely that an analyst’s observable
optimism and the strength of his beliefs are correlated, especially for IPO firms, about
which there is little data other than a priori beliefs upon which to base an assessment.
Second, there is no reason to rule out the possibility that managers do select underwriters
based on the strength of beliefs. Managers of issuing firms typically meet in person with
the analysts employed by prospective issuers, at which point they have every opportunity
to ascertain the strength of their beliefs.
Bradshaw, Richardson, and Sloan (2003) conduct a study in which the selection
bias hypothesis mentioned is a less compelling competing explanation to their results
which, at first glance, seem to support the conflict of interest hypothesis. Their results do
not support the conflict of interest hypothesis, however, because there exists yet another
competing hypothesis involving another type of selection bias, which can explain their
findings.
Instead of comparing the forecasts and recommendations of affiliated and
unaffiliated analysts, Bradshaw, Richardson, and Sloan (BRS) compare the consensus
forecasts and recommendations of firms that are net issuers of securities with firms that
are not and find that the former are more optimistic than the latter. Since many
investment banks may be competing for the financing business of a given firm, BRS
11
argue that conflict of interest bias may be present in the reports of unaffiliated analysts as
such analysts may bias their forecasts in an attempt to help their employers win
investment banking business, even though they ultimately fail. Thus BRS construe their
finding of more optimistic consensus forecasts and recommendations of analysts
reporting on firms that are about to issue or have just issued securities as evidence in
favor or the conflict of interest hypothesis. The selection bias hypothesis as outlined
above is a less compelling explanation of these results because the reports of analysts
whose employers were not selected as underwriters are included in the consensus. It is
important to note, however, that this methodology only reduces and does not eliminate
the selection bias effect since affiliated analysts’ reports are also part of the consensus,
which for most firms includes no more than two or three analysts.
The BRS methodology introduces another type of selection bias. McNichols and
O’Brien (1997) find that those analysts who are less optimistic about a firm are less likely
to cover it. Thus the distribution of analyst forecasts and recommendations is censored
on the left, thereby making it appear that analysts on average are optimistically biased. It
is reasonable to conjecture that this phenomenon, henceforth the “self-selection
phenomenon,” is less pronounced for larger, better-established firms. Analysts less
optimistic about a given firm may be more likely to cover it if it is larger and better
established. Firms issuing equity tend to be smaller, and less well established, so finding
greater optimism in analyst reports on firms issuing equity could be a result of the self-
selection phenomenon rather than conflict of interest. The self-selection phenomenon,
coupled with major statistical biases inherent to BRS’s methodology, makes it drawing
12
any conclusions about the validity of the conflict of hypothesis from their results
problematic.10
2.4.2 Conflict of Interest in the M&A Context
Since mangers like optimistic analyst coverage of their firms, the conflict of
interest hypothesis would predict analysts affiliated with acquirer advisors, in both cash
and stock deals, would tend to be optimistic about acquirer stock. If analysts can be
swayed by investment banking business, the promise of repeat buy-side M&A business
would seem to be an effective means for managers to obtain optimistic coverage, since
many firms make multiple acquisitions during their lifetimes. Since the managers of
acquiring firms seek to purchase the target at as low a price as possible, the conflict of
interest hypothesis also predicts that analysts affiliated with the acquirer advisor, in both
cash and stock deals, would tend to be pessimistic about target stock.
Since target managers want to obtain as high a price as possible, the conflict of
interest hypothesis predicts that analysts affiliated with target advisors, in both cash and
stock deals, will be optimistic about the target. In cash deals, the acquirer’s stock is
irrelevant to the target managers & shareholders, so the conflict of interest hypothesis
predicts neither optimism nor pessimism on the part of target-affiliated analysts reporting
on acquirer stock. In stock deals, however, target shareholders and managers do have an
interest in acquirer stock. Since they will want as many acquirer shares exchanged for
10 BRS introduce statistical biases in their calculation of the optimism of analyst long-term EPS growth forecasts. First, in calculating the benchmark long-term growth rate used to estimate analyst optimism, they implicitly assume that analyst growth forecasts are continuously compounding rates. If analyst forecasts are of an annually compounding rate, a more reasonable assumption, BRS severely underestimate the benchmark rate and hence overestimate analyst optimism. Further, this overestimate is higher for high-growth firms, i.e., those more likely to issue securities, than low growth firms. Second, they calculate the benchmark long-term growth rate by using a log transformation of EPS, which also causes them to overestimate analyst optimism, and this overestimate is also higher for high-growth firms. Together, these biases can explain much of their results related to long-term growth forecasts.
13
target shares as possible, before the transaction they will want pessimistic coverage of the
acquirer. Since immediately after the transaction target shareholders and managers
become acquirer shareholders, they will want optimistic coverage of the acquirer. Hence
in stock deals, the conflict of interest hypothesis predicts pessimism before the
transaction and optimism after the transaction on the part of target-affiliated analysts
reporting on the acquirer. Since the target no longer exists after the transaction, and
hence cannot offer any repeat business to the target advisor, one might argue that the
target advisor has no incentive to pressure its analyst into favorable coverage of the
acquirer after the transaction. However, if target shareholders like positive analyst
coverage of acquirer stock after a stock transaction, and a target advisor makes it a policy
to provide such coverage, it will be more likely to be selected as an advisor by other
targets in the future.
2.4.3 Selection Bias in the M&A Context
To understand how selection bias may work in the M&A context, it is necessary
to outline the duties of M&A advisors and the various regulations involved. Our
discussion here is based on conversations with practitioners and the personal experience
of one of the authors. Target and acquirer advisors are typically hired some time before
the transaction is announced. In most friendly deals, negotiations between target and
acquirer begin before the transaction announcement as well. The target advisor’s job is
to get as high a price for target stock as possible, and in a stock deal it also involves
making a case for as low a valuation of acquirer stock as possible. The lower the
valuation of acquirer stock, the greater the number of acquirer shares exchanged for
target shares. The acquirer advisor’s job is the opposite: to argue for as low a valuation
14
as possible for the target stock, and in a stock deal to also argue for as high a valuation as
possible for the acquirer stock. Both acquirer and target advisors typically must also
convince their clients’ boards that the terms of the deal are satisfactory. Analysts
affiliated with either advisor are free to issue reports on both target and acquirer stock so
long as they are kept out of the M&A process and given no inside information. These
conditions are nearly always met.
To the extent that the opinion of an analyst in an advisor’s employ affects the
latter’s ability to make the case for a low or high valuation of a given stock, and hence
the advisor’s execution ability, one would expect analyst opinion to influence a
prospective M&A client’s choice of advisor. Hence selection bias predicts that target
advisor-affiliated analysts would tend to be optimistic about target stock. By a similar
argument, selection bias predicts that an acquirer advisor in a stock deal would tend to
have optimistic analysts in his employ. However, the selection bias hypothesis predicts
that acquirer advisor-affiliated analysts will be neither pessimistic nor optimistic since in
cash deals the acquirer stock is irrelevant. If employing an analyst pessimistic about the
target makes it easier for the acquirer advisor to argue for a low target valuation, the
selection bias hypothesis predicts that analysts affiliated with acquirer advisors will be
pessimistic about target stock. Finally, in stock deals, the target advisor must make the
case for a low acquirer valuation, so selection bias predicts that analysts affiliated with
target advisors will tend to be pessimistic about the acquirer in stock deals. In cash deals,
the acquirer’s stock is irrelevant, so the hypothesis makes no predictions of target
advisor-affiliated analyst pessimism or optimism about acquirer stock.
15
In addition, there seems to be less of a reason for selection bias in M&A
relationships than in equity underwriting relationships. Legally, investment banks are
obliged to keep their research activity separate from their M&A and underwriting
activity. This institutional separation is often referred to as the “Chinese Wall.” If an
analyst in some way becomes involved in an M&A or underwriting transaction, he is
prohibited from writing reports about parties to the transaction while involved in it and
for a period afterward. In nearly every equity underwriting deal, analysts are “brought
over the wall” and heavily involved. Their participation in the underwriter’s sales pitch
to the public is essential to the success of the deal. Krigman, Shaw, and Womack (2001)
find that the promise of quality analyst coverage heavily influences underwriter choice.
Since the analyst is so heavily involved in selling the deal, it seems that an underwriter
with more optimistic analysts will be better able to execute a stock issuance, and hence
selection bias should be significant in the underwriting context.
In the case of M&A transactions, however, analysts are seldom if ever “brought
over the wall.” Since the analyst is not in any way involved in the M&A advisor’s
attempts to make the case for a low or high valuation of an acquirer’s or target’s stock, it
does not seem likely that the analyst opinion has a large effect on the advisor’s ability to
make such a case. Furthermore, the case for a low or high valuation typically has to be
made to a board, and not the public, but boards have inside information about all firms
involved that no analyst has access to, so it seems unlikely that boards would be
influenced by analyst opinion. Of course, the board does eventually have to make the
case for the deal to the public, which is influenced by analyst opinion, so some potential
16
for selection bias, albeit more muted than in the underwriting context, still exists in the
M&A context.
4.2.4 Where the Two Hypotheses Make Different Predictions
As we have seen, the conflict of interest hypothesis predicts that acquirer advisor-
affiliated analysts will be optimistic, but the selection bias hypothesis predicts neither
optimism nor pessimism. Before a stock transaction, both hypotheses predict target
advisor-affiliated analysts will be pessimistic about acquirer stock, but after the
transaction the conflict of interest hypothesis predicts optimism and the selection bias
hypothesis predicts pessimism. By examining the whether analysts are optimistic or
pessimistic in these scenarios, we can help distinguish between the hypotheses. Tables 1
presents all the different scenarios and what one might predict about analyst optimism or
pessimism under the conflict of interest and selection bias hypotheses.
3. Data and Descriptive Statistics
We obtain M&A transaction data from SDC for years 1993 to 2001. We exclude
from our sample buybacks, acquisitions of partial interest, recapitalizations, spin-offs,
split offs, exchange offers, and acquisitions of remaining interest because analyst
incentives in such deals are unclear. Thus our sample consists solely of statutory
mergers, acquisitions of assets, and acquisitions of certain assets.11 We also limit the
sample to deals in which either the target or acquirer or both are public, as deals for
which fee information is available, and deals which are completed. The number of deals
11 To implement our sample selection, include only SDC deals in which the field “form of deal” is labeled as ‘AA’, ‘AC’, or ‘M.’ These labels correspond to “acquisit ion of assets,” “acquisition of certain assets,” and “statutory merger”, respectively. This method of sample selection is the same as excluding deals whose “form of deal” field is labeled ‘A’, ‘AR’, ‘AP’, ‘R’, ‘B’, and ‘EO,’ which correspond to spin-offs, acquisitions of remaining interest, acquisitions of partial interest, recapitalizations, buybacks, and exchange offers.
17
comes to 2,922. In 1,713 of these deals the acquirer paid more than 50% of the
acquisition price in stock. In 1,201 the acquirer paid 50% or more in cash or other non-
stock currency (usually assumed debt). In 18 deals no information was provided on
currency. 44 of the deals were acquisitions of assts, 2887 were statutory mergers, and 1
was an acquisition of certain assets.
From I/B/E/S we obtain all available 1YR and 2YR-ahead EPS forecasts, long-
term growth forecasts, and recommendations for all acquirers, targets, and their
immediate and ultimate parents (as defined by SDC) published within one year of each
transaction in our sample. We restrict our attention to the above forecasts because they
are widely available for nearly every firm in our sample. Forecasts for horizons longer
than 2 years are available for very few firms. In addition we obtain from I/B/E/S the
closing price and shares outstanding of the stock on the last trading day of the calendar
month in which the forecast was published. If a price for this day is not available, we take
the closing price on the day closet to it, provided it is within 30 days. Table 2 presents
descriptive statistics on the transaction values of the deals for which at least one I/B/E/S
forecast or recommendation is available.
Next we determine the affiliation of each analyst who issued a forecast or
recommendation in our sample. This task is not complicated in principle because SDC
lists all M&A advisors retained on a deal, and I/B/E/S provides the name of the securities
firm, which it calls the “broker,” employing each analyst issuing a forecast or
recommendation. Unfortunately, the SDC codes for M&A advisors and I/B/E/S codes
for brokers are different, and there is no mapping between the two coding systems.
Hence we must individually match I/B/E/S brokers and SDC advisors by hand using their
18
corporate names. In many instances, the names in the two databases are qualitatively the
same and can be matched by sight.
In many other instances, however, the broker listed in the I/B/E/S database may
be a subsidiary of an advisor in the SDC database, or vice-versa, and the names of the
broker and advisor bear no similarities. In some instances, the SDC advisor and I/B/E/S
broker are subsidiaries, with completely different names, of the same parent company.
To check of such affiliations, we look up each I/B/E/S broker in Hoovers Online, the
Directory of Corporate Affiliations, as well as Lexus-Nexus and corporate webpages to
see if it has subsidiary-parent or common parent affiliations with one of the SDC advisors
in our sample. We also look up each SDC advisor in our sample. Through this method,
we are able to detect subsidiary-parent and common parent affiliations that continue until
the present. Unfortunately, we have no way of detecting affiliations that were terminated
in the past, unless there were a news stories about them. Still, since our sample begins
only in 1993, this problem should not create too much concern.
In all of our tests, we seek to determine whether forecasts and recommendations
issued by analysts affiliated with M&A advisors (“affiliated analysts”) are optimistic or
pessimistic relative to consensus. Because we do not want any forecast in our consensus
to be contaminated by M&A affiliation, we exclude from consensus any forecasts or
recommendations issued by analysts affiliated with an M&A advisor that was retained
within one year of the forecast date by either the firm whose EPS is being forecasted or
stock is being recommended or the counterparty to the M&A transaction.
Recent research indicates that herding behavior may be economically significant.
Scharfstein and Stein (1990) initiated the herding literature with their model of firm
19
manager herding, which Trueman (1994) applies to analysts. Hong and Kubik (2000) as
well as Welch (2000) find evidence that analysts do indeed exhibit herding behavior.
Hence to make sure that herding by unaffiliated analysts does not taint our estimate of the
unaffiliated consensus, we exclude from our consensus estimate any forecast or
recommendation issued after the one issued by the affiliated analyst. In the case of
forecasts, to calculate the consensus, we average all the unaffiliated forecasts and/or
recommendations for a given firm issued within a calendar month and before the
affiliated analyst’s. We then calculate the difference between the affiliated analyst’s
forecast or recommendation and the consensus. In the case of EPS forecasts, we
normalize the difference by the closing price as of the end of the month. If there is no
closing price for the last trading day of the month, we use the closing price on the day
closest to the last trading day for which one is available, provided this day is less than 30
days away from the last trading day. We also calculate each unaffiliated analyst’s
deviation from consensus in the same manner, except we define consensus in this case as
the average of all other unaffiliated analysts’ recommendations or forecasts.
Table 3 presents the scaled average deviation from consensus of affiliated analyst
one-year-ahead EPS forecasts, 2-year ahead EPS forecasts, as well as absolute deviation
from consensus of long-term growth forcasts and recommendations. The data are sorted
by the currency use in the deal, analyst affiliation, the target or acquirer status of the firm
upon which analyst is reporting, as well as whether the report was issued before or after
the M&A transaction. We classify as stock deals those in which 50% or more of the
consideration was paid in stock, and others we classify as cash.
20
The evidence in the table shows there to be little, if any, association between
analyst affiliation in and relative optimism in forecasts. In nearly all cases, the mean
optimism of affiliated analysts is statistically indistinguishable from zero. In those few
cases where the mean is statistically significant, the sign is usually contrary to that
predicted by the conflict of interest hypothesis. In the case of recommendations, the
deviation from consensus is significant in the direction of the conflict of interest
hypothesis in 5 out of 16 instances, but in none of those instances is it economically
significant as the deviation is always less than 0.3 percentage points. Hence this analysis
gives us very little, if any, evidence in favor of the conflict of interest hypothesis.
Thus far, we find little evidence of an association between M&A relationships
and analyst optimism or pessimism relative to consensus. While the analysis of mean
optimism/pessimism is instructive, it fails to take into account the numerous factors, such
as investment banking fees, which might affect the degree to which an M&A relationship
will affect analyst objectivity. It also has the limitation of treating all variables in a
binary fashion. For instance, we either classify a deal as a cash deal or stock deal, yet in
many deals the acquirer pays for the target with some combination of stock and cash. In
the next section we will use regression analysis in order to address such issues.
4. Ordinary Least Squares Analysis of Forecasts
In this section we use regression analysis to determine whether there is an
association between M&A relationships and the relative optimism of EPS and long-term
growth forecasts of analysts affiliated with M&A advisors.
For each forecast or recommendation, let Da equal 1 if the analyst is affiliated
with the acquirer and zero otherwise. Likewise, let Dt equal 1 if the analyst is affiliated
21
with the target and zero otherwise. Finally, let A equal 1 if it is the acquirer’s EPS or
growth that is being forecast or recommended, respectively, and 0 if it is the target’s.
Since our hypotheses depend on both whose EPS or growth is being forecast or
recommended as well as the affiliation of the analyst, it is useful to define the following
row vector, V, of interaction terms:
V = < ADt,, (1-A)Dt, ADa, (1-A)Da>
To see how investment banking fees affect analyst forecasts, we run the following
regression for EPS and long-term growth forecasts:
O = α +AFa β1 +(1-A)Fa β2 + AFt β3 + (1-A)Ft β4 + (365-d)Va + LVb + MVc + NVc+ε (Model 1) where O is the analyst’s relative optimism, defined as the analyst’s forecast less the
consensus forecast. In the case of EPS forecasts, we normalize this difference by the
month-end stock price. Ft is the dollar amount of fees paid to the analyst’s employer by
the target. Fa is the dollar amount of fees paid to the analyst’s employer by the acquirer.
β1, β2, β3, and β4 are regression coefficients. We interact the fee variables with A and
1-A because, based on our hypotheses, we make predictions according to who is paying
fees as well as whose EPS or long-term growth is being forecast. For instance, it follows
from both the conflict of interest and selection bias hypotheses that analysts receiving
fees from the acquirer should be more optimistic about acquirer EPS or growth, and
therefore we would predict that β1 would be positive based on these hypotheses.
Likewise, based on these hypotheses, we would predict β4 to be positive, β2 to be
negative, and the predicted sign on β3 is ambiguous.
a, b, c, and d are column vectors of regression coefficients corresponding to the
interactions of V with our control variables, which we shall now describe. d is the
22
absolute number of days between the transaction and the date on which the analyst
forecast was issued. We include (365-d) interacted with V as a control because affiliated
analyst forecasts, if they are biased, are more likely to be biased closer to the transaction
date. L is the length of the analysts career, that is, the difference in years between the
date of the forecast or recommendation date in our sample and the date of the analyst’s
first ever forecast or recommendation recorded in the I/B/E/S database. We include L
because recent research shows that more experienced analysts are more likely to deviate
from consensus (Hong, Kubik, and Solomon, 2000). M is the market capitalization of the
firm as of the end of the calendar month. We include it because uncertainty about future
earnings is likely lower or larger firms. Finally, N is the number of analysts following
the company, measured as the total number of analysts who issued at least one forecast
within the same quarter. We include N because a greater number of analysts increases
the likelihood of herding and therefore lessens dispersion in analyst forecasts. We
interact all our control variables with V because they influence the variance, rather than
the mean of the distribution of analyst forecasts. Thus these control variables are only
going to affect the expected value of a given analyst’s deviation from the consensus (or
mean) if we already have a prior as to whether his forecast is likely to be greater or
smaller than consensus. In our analysis, our priors are based on affiliation as well as
which firm’s EPS or growth is being estimated, so we interact our variance-affecting
control variables with V.
In order to estimate model 1, we take the Cartesian product of every analyst
forecast with every M&A transaction in our database to which a firm whose EPS or
growth is being forecast was a party. Thus we obtain as observations every forecast-
23
transaction pair possible. Note that the same forecast may appear in multiple
observations since some firms engage in multiple M&A transactions during the course of
a year. We only include in our sample EPS or growth forecasts or recommendations of
targets and acquirers, excluding their parents. We exclude from consensus, however,
forecasts and recommendations of analysts who are affiliated with the parent of a
company that is being reported on.
In order to ensure that cross-sectional error correlation does not bias our standard
error estimates, we estimate model 1 using quarterly Fama-MacBeth ordinary least
squares regressions. The following describes this procedure in detail. First we sort the
data by calendar quarter and estimate the parameters for each quarter using ordinary least
squares. We then compute parameter estimates for the whole sample by computing the
sample means of the quarterly estimates, weighted by the number of observations in each
quarter. We also compute standard error estimates and p-values using this quarterly time-
series of parameter estimates, also weighting by number of observations in each quarter.
Our estimates of the parameters of model 1 can be found in Table 4, in the
columns labeled “model 1.” Also shown are standard errors and p-values. In casees
where the conflict of interest hypothesis provides a prediction for the sign of a
coefficient, the p-values are one-sided. None of the coefficient estimates in the “model
1” columns of the table is statistically significant in the predicted direction, even at the
10% level. Thus, based on our results for model 1, we cannot reject the null hypothesis
that there is no relationship between analyst optimism and investment baking fees.
Furthermore, as demonstrated by the standard errors, our power to reject this null is high.
Where we use analyst optimism in EPS forecasts as the dependent variable, the standard
24
errors are so small as to be economically negligible. Where we use analyst optimism in
long-term growth forecasts as the dependent variable, our standard errors, with one
exception, are all less than 1. With a standard error of 1, affiliated analysts’ deviation
from consensus for every $1 million received in investment banking fees could be as low
as 1.2 percentage points, and we would still be able to detect it at the 10% level of
significance using a one-sided test. The only coefficient whose standard error estimate is
greater than 1 is the coefficient on (1-A)Ft, whose standard error equals 1.63. Even with
such a high standard error, however, we could detect an affiliated analyst deviation from
consensus as low as 2 percentage points per $1 million in investment banking fees using
a one-sided test at the 10% significance level. While such a deviation from consensus is
not negligible, is still low compared to most growth forecasts. Hence even for the
coefficient with the comparatively large standard error estimate, it appears we have high
power to reject the null that affiliated analysts do not deviate from consensus.
The statistical tests based on model 1, while valid tests of the association between
investment banking fees and analyst optimism in forecasts, do not discriminate between
the conflict of interest and selection bias hypotheses discussed in the previous section.
Since we fail to reject the null of no association, however, it does not matter that we fail
to distinguish between hypotheses since the result of no association is consistent with
neither hypothesis. Still, it is useful to test the hypotheses separately, and for this purpose
we develop model 2, which we discuss next.
The type of consideration paid may alter the incentives of affiliated analysts, and
thus we use a measure of it to help distinguish between the selection bias and conflict of
interest bias hypotheses. We define two interaction terms:
25
Itb = BASFt
Ita = (1-B)ASFt
Where B is a dummy equal to 1 if the forecast was issued before the transaction
announcement date and zero otherwise, and S is the percentage of consideration paid for
in stock. These two interaction terms seek to measure the association the relative
optimism in the target-affiliated analyst's estimate of acquirer EPS between and 1) target
fees paid and 2) the percentage of the consideration paid in stock. Itb measures this
association when the estimate date falls before the announcement date, Ita measures it
when the estimate date falls after the announcement date. As explained in Section 2,
under the selection bias hypothesis, analysts affiliated with the target in a stock deal issue
pessimistic forecasts of acquirer EPS and growth, so under this hypothesis the coefficient
on both these interaction terms should be negative. In Section 2 we also saw that under
the conflict of interest hypothesis, analysts affiliated with the target advisor in a stock
deal should be optimistic about acquirer EPS and growth, at least after the transaction, so
the coefficient on Ita should be positive. If both selection bias and conflict of interest
effects are present, the coefficient on Itb should be less than Ita. If conflict of interest
hypothesis dominates, the coefficient on Itb should be negative and the coefficient on Ita
positive.
Finally, define a third interaction term:
Iac = A(1-S)Fa
This interaction term measures the association between the acquirer-affiliated analyst's
bias in estimating acquirer EPS and growth and the percentage of consideration paid in a
manner other than acquirer stock as well as the fees paid by the acquirer. Recall from
Section 2 that there should be no selection bias in forecasts of acquirer EPS issued by
26
acquirer-affiliated analysts in cash deals. Hence if conflict of interest does not affect
analyst objectivity, the coefficient on this term should be zero. If conflict of interest
problems exist and managers of acquiring firms reward analysts for favorable coverage
with M&A business, it should be positive. Selection bias should not affect the
coefficient, for reasons given in section 2. We thus have another regression specification:
O = α +β1AFa + β2(1-A)Fa + β3AFt + β4(1-A)Ft + (365-d)Va + LVb + MVc + NVc + + γ1Iac + γ2Itb + γ3Ita +ε (Model 2)
where γ1 , γ2 , and γ3 are regression coefficients.
As with model 1, we estimate model 2 by running quarterly Fama-McBeth type
ordinary least squares regressions. Our estimates of the parameters of model 2 are
presented in Table 4 in the columns labeled “model 2,” along with standard errors and p-
values. P-values are one-sided where the conflict of interest hypothesis predicts the sign
of a parameter. None of our estimates is statistically significant at the 10% level or
better. Hence we cannot reject the null of no conflict of interest bias. As before, our
power to reject is high in most cases. Where optimism in EPS forecasts is the dependent
variable, the standard errors are economically negligible.
In the case where optimism in long-term growth forecasts is the dependent
variable, the standard errors are low for all coefficient estimates except for those of terms
interacting target fees and A, the dummy indicating that the acquirer’s growth is being
forecast. The high standard error in these cases is driven by the fact that we have four
variables interacted with target fees, and these four variables are likely to be highly
correlated.
A potential problem with our analysis lies with the possibility that the relationship
between affiliated analyst deviation from consensus and investment banking fees may be
27
nonlinear. In order to test for this possibility, for each year we rank target and acquirer
fees and compute fee terciles. We define a variable, t_feerank, which takes the value of
the tercile to which the fees received from the target belong. We define a second
variable, a_feerank, which takes the value of the tercile to which the fees received from
the acquirer belong. a_feerank or t_feerank equal zero if the analyst is unaffiliated with
the acquirer or target, respectively. We re-estimate models 1 and 2, substituting
a_feerank and t_feerank for Fa and Ft. The results for this specification are qualitatively
similar to the results from the linear one, except that the standard are higher, indicating
that the linear specification is probably more appropriate. For reasons of brevity, we do
not report the results from the non-linear specification. To give the reader a sense of how
little difference the magnitude of fees makes, in Figures 3a-3c we report selected
descriptive statistics for deviation from consensus EPS and growth forecasts of affiliated
analysts within the high and low fee terciles.
5. Ordered Logistic Regression Analysis for Recommendations
In this section we use ordered logistic regression analysis to determine whether
M&A relationships affect the relative optimism of analyst recommendations. We use
similar specifications to those we used in section 4 to test our hypotheses.
Recommendations, however, unlike forecasts, are discrete and ordinal. They are
assigned a value of 0 if the analyst issues a strong-sell recommendation and are assigned
a value of 4 if the analyst issues a strong buy. Buy, hold, and sell recommendations are
assigned values of 3,2, and 1, respectively.
The discrete and ordinal nature of recommendations makes the ordinary least
squares regression in which recommendations are the dependent variable an inappropriate
28
statistical tool, since it will produce biased and inconsistent parameter estimates.
Similarly, the discrete and ordinal nature of recommendations makes deviation from
consensus a suspect measure of analyst optimism. We therefore use a logistic
specification because it is naturally better suited to dealing with discrete dependent
variables.
The dependent variable in all of our logistic regressions is the analyst
recommendation. In order assure that the estimates of the coefficients on our
independent variables are valid measures of those variables’ impact on analyst optimism
relative to consensus, we also include as independent variables the number of strong buy,
buy, hold, sell and strong sell recommendations previously issued by unaffiliated
analysts. Let U be a vector that contains these 5 variables. The 3 logit models we
estimate are described below:
Logit Model Independent variables
1 U
2 U, AFa, (1-A)Fa, AFt, (1-A)Ft, (365-d)V, LV, MV, NV 3 U, AFa, (1-A)Fa, AFt, (1-A)Ft, (365-d)V, LV, MV, NV, Iac, Itb, and Ita
Logit models 2 and 3, in addition to U, contain the same independent variables as the
OLS model (1) and (2) in Section 4. We include logistic model 1 to estimate the extent
to which an analyst’s unaffiliated colleagues’ recommendations explain his own
recommendations, and hence to verify that the inclusion of U as an independent variable
causes our estimates of other coefficients to be valid measures of the extent to which
29
other factors influence analyst optimism relative to consensus, rather than absolute
optimism.
Unfortunately, we cannot estimate our logistic models using Fama-McBeth type
methodology because in some calendar quarters there are too few observations to
compute parameter estimates, so we estimate our logistic models using the full panel of
data. If error cross-correlations exist in the data, therefore, our standard error estimates
will be biased toward zero, making us more likely to erroneously reject the null
hypothesis of no conflict or interest. The fact that we still fail to reject the null despite
the potential bias, however, only strengthens our results.
The results of our estimates of logistic models 1-3 can be found in Table 5. We
report odds ratios estimates along with the standard error of the odds ratio and the p-
value. For reasons of spatial parsimony, we do not report results for the control variables,
which are available upon request. We estimate the odds ratio standard error using the
delta method, and we estimate p-values using the logit parameter estimates and their
standard errors, which we do not report since their economic interpretation is not as
intuitive as that of the odds ratio. Since the ratio of a logit parameter estimate to its
standard error, under reasonable assumptions, is Gaussian, we use a Gaussian, rather than
t, CDF to compute p-values.
In all the models, the only statistically significant odds ratios are those
corresponding to U, the vector of recommendation frequencies of unaffiliated colleagues.
All others are not significant at even the 10% level, so we cannot reject the null
hypothesis that there is no association between M&A fees and analyst pessimism and/or
optimism. As with model 2 in section 4, model 3 in this section allows us to distinguish
30
between the conflict of interest hypothesis and selection bias hypothesis. We fail to
reject the null that neither of these biases is present.
Given the small size of the standard errors, in all the models, our power to reject
the null is high. The highest standard error is 4.3%, meaning that if $1 million in
investment banking fees were to make analysts at least (approximately) 5.6% more likely
to give a higher (or lower) recommendation, we would be able to detect it at the 10%
level of significance using a one-sided test. While not negligible, 5.6% is not
economically large. Hence our results provide powerful evidence that investment
banking fees fail to influence analyst recommendations.
As with our analysis of forecasts, our analysis of recommendations may suffer
from the possibility that the relationship between investment banking fees and the
probability of an analyst deviating from consensus is non- linear. As with the OLS
models in section 4, we re-estimate all the models in this section substituting a_feerank
and t_feerank for Fa and Ft, respectively. The results are qualitatively similar, except that
the standard errors of the parameter estimates of the new specification are higher. This
indicates that the new specification is inferior, so for reasons of brevity we do not report
its results. To give the reader a sense of how little difference the magnitude of fees
makes, in Figure 3d we report selected descriptive statistics for deviation from the
consensus recommendation for affiliated analysts within the high and low fee terciles.
6. Summary and Conclusions
Previous research has documented that analysts affiliated with equity underwriters
tend to issue reports on client firms that are more optimistic than the reports issued by
unaffiliated analysts. Some researchers claim that these results constitute evidence in
31
favor of what we call the conflict of interest hypothesis: the hypothesis that affiliated
analysts deliberately bias their reports on client firms. There is, however, an alternative
hypothesis, which we call the selection bias hypothesis, that can explain these results. It
is possible that affiliated analysts are objective, but that underwriters whose analysts are
more optimistic about a given firm are more likely to underwrite that firm’s equity
issuance.
Our finding that there is no association between M&A advisory relationships and
analyst optimism or pessimism constitutes strong evidence against the conflict of interest
hypothesis ; our results suggest that the selection bias hypothesis is the more likely
explanation of equity underwriter-affiliated analyst optimism. The incentives analyst
face to bias reports on M&A clients are as least as large as those they face to bias reports
on equity underwriting clients. M&A fees constitute at least as great a share of
investment banking revenues as do underwriting fees, and firms engaging in M&A
transactions have no less a reason to want favorable analyst coverage than firms engaging
in equity underwriting transactions. Furthermore, as we have demonstrated, selection
bias is unlikely to be present in the M&A context. In fact, there are instances in which it
is impossible, such as in the case of an analyst affiliated with an acquirer in an all-cash
transaction. Thus our finding that M&A fees do not influence analyst forecasts and
recommendations constitutes strong evidence that any appearance of underwriting fees
doing so is due mainly to selection bias.
Our results, however, do not necessarily indicate that there are no potential
conflict of interest problems. As mentioned previously, “Chinese Wall” regulations
32
restrain communication between investment bankers and analysts, so it is possible that
conflicts of interest exist but are being successfully thwarted.
33
References
Bradshaw, Mark T., Scott A. Richardson and Richard G. Sloan. “Pump and Dump: An Empirical Analysis of the Relation Between Corporate Financing Activities and Sell-Side Analyst Research.” Working Paper, University of Michigan Business School, 2003.
Dugar, Amitabh and Siva Nathan. “The Effect of Investment Banking Relationships on Financial Analysts’ Earnings Forecasts and Investment Recommendations.” Contemporary Accounting Research 12 (1995): pp. 131-160.
Dechow, Patricia M., Amy P. Hutton and Richard G. Sloan. “The Relation Between Analysts’ Forecasts of Long-Term Earnings Growth and Stock Performance Following Equity Offerings.” Contemporary Accounting Research 17 (2000): pp. 1-32.
Krigman, Laurie, Wayne H. Shaw, and Kent Womack. “Why Do Firms Switch Underwriters?” Journal of Financial Economics 60 (2001): 245-84
Hunter, William C. and Mary Beth Walker. “An Empirical Examination of Investment Banking Merger Fee Contracts.” Southern Economic Journal 56 (1990): pp, 1117-1130.
Lin, Hsiou-wie and Maureen F. McNIchols. “Underwriting Relationships, Analysts’ Earnings Forecasts and Investment Recommendations.” Journal of Accounting and Economics 25 (1998): 101-127.
Lin, Hsiou-wie, Maureen F. McNIchols and Patricia O’Brien. “Analyst Impartiality and Investment Banking Relationships.” Unpublished Working Paper, 2003.
McNichols, Maureen and Patricia C. O’Brien. “Self-Selection and Analyst Coverage.” Journal of Accounting Research 35 (1997): pp. 167-199.
Michaely, Roni and Kent Womach. “Conflict of Interest and the Credibility of Underwriter Analyst Recommendations.” Review of Financial Studies 12 (1999): 653-686.
Hong, Harrison, Jeffery Kubik and Amit Solomon. “Security analysts’ career concerns and herding of earnings forecasts.” Rand Journal of Economics 31 (2000): pp. 121-144.
Servaes, Henri and Marc Zenner. “The Role of Investment Banks in Acquisitions.” Review of Financial Studies 9 (1996): 787-815.
Scharfstein, David and Jeremy Stein. “Herd Behavior and Investment.” American Economic Review 80 (1990): pp. 465-479.
34
Trueman. “Analyst forecasts and herding behavior.” Review of Financial Studies 7 (1994): pp. 97-124.
Welch, I. “Herding Among Securities Analysts.” Journal of Financial Economics 58 (2000): 369-396.
Womack, Kent. “Do Brokerage Analysts’ Recommendations Have Investment Value?” Journal of Finance 51 (1996): 137-167.
35
Figure 1: Aggregate US Investment Banking Fee Revenues ($Millions)
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
1994 1995 1996 1997 1998 1999 2000 2001 2002
Equity Underwriting
M&A Advisory
Source: Freeman & Co. estimates. Freeman & Co. compiles its estimates using Captial IQ, Bloomberg, Thomson Financial, and SDC
36
Figure 2: Market Capitalization of Firms Issuing Equity vs. Firms Engaging in M&A Activityas a % of Aggregate Equity Market Capitalization
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
Firms Issuing Equity
Firms Engaging in M&A Activity
37
Figure 3a: Deviation From Consensus 1-Year Ahead EPS Forecasts of Affiliated Analysts Paid High and Low Fees
(0.02) (0.10)
(0.41)
(0.11)
(0.00)
(0.09)
(0.12)
(0.15)
(0.46)
(0.45)
-0.0045
-0.004
-0.0035
-0.003
-0.0025
-0.002
-0.0015
-0.001
-0.0005
0
0.0005
0.001
Stock Deals Affiliation:Target,Subject: Acquirer,
Before Deal
Stock Deals Affiliation:Target,Subject: Acquirer,
After Deal
Affiliation: Target,Subject: Target
Affiliation: Acquirer,Subject: Acquirer
Affiliation: Acquirer,Subject: Target
Fees in top 1/3 Fees in Bottom 1/3 (p-values)
38
Figure 3b: Deviation From Consensus 2-Year Ahead EPS Forecasts of Affiliated Analysts Paid High and Low Fees
(0.94)
(0.46) (0.53)
(0.89)
(0.48)
(0.02)
(0.04)
(0.46) (0.82)
(0.45)
-0.003
-0.0025
-0.002
-0.0015
-0.001
-0.0005
0
0.0005
0.001
Stock Deals Affiliation:Target,Subject: Acquirer,
Before Deal
Stock Deals Affiliation:Target,Subject: Acquirer,
After Deal
Affiliation: Target,Subject: Target
Affiliation: Acquirer,Subject: Acquirer
Affiliation: Acquirer,Subject: Target
Fees in top 1/3 Fees in Bottom 1/3 (p-values)
39
Figure 3c: Deviation From Consensus Long-Term Growth Forecasts of Affil iated Analysts Paid High and Low Fees
(0.91)
(0.16)
(0.93)
(0.42)
(0.07)
(0.58)
(0.25)
(0.70)(0.58)
(0.69)
-3
-2
-1
0
1
2
3
Stock Deals Affiliation:Target,Subject: Acquirer,
Before Deal
Stock Deals Affiliation:Target,Subject: Acquirer,
After Deal
Affiliation: Target,Subject: Target
Affiliation: Acquirer,Subject: Acquirer
Affiliation: Acquirer,Subject: Target
Fees in top 1/3 Fees in Bottom 1/3 (p-values)
40
Figure 3d: Deviation From Consensus Recommendation of Affi l iated Analysts Paid High and Low Fees
(0.02)
(0.30)(0.34)(0.37)(0.29)
(0.00)
(0.16)(0.22)
(0.55)
(0.40)
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Stock Deals Affiliation:Target,Subject: Acquirer,
Before Deal
Stock Deals Affiliation:Target,Subject: Acquirer,
After Deal
Affiliation: Target,Subject: Target
Affiliation: Acquirer,Subject: Acquirer
Affiliation: Acquirer,Subject: Target
Fees in top 1/3 Fees in Bottom 1/3 (p-values)
41
Table 1: Predictions Under Selection Bias and Conflict of Interest Hypotheses
Panel A: Stock Deal Predictions
Subject of Report and Hypothesis Affiliation and Time of Analyst Report (Before or After
Transaction)
Target: Selection Bias
Target: Conflict of
Interest
Acquirer: Selection Bias
Acquirer: Conflict of
Interest Target: Before Positive Positive Negative Negative Target: After Positive Positive Negative Positive Acquirer: Before Negative Negative Positive Positive Acquirer: After Negative Negative Positive Positive Predictions of Analyst Bias Under the Selection Bias and Conflict Of Interest Hypotheses in Stock Deals. Predictions are sorted by analyst affiliation, whether the reports are on the acquirer or target, and whether the report is issued before or after the transaction.
Panel B: Cash Deal Predictions Subject of Report and Hypothesis Affiliation and Time
of Analyst Report (Before or After
Transaction)
Target: Selection Bias
Target: Conflict of
Interest
Acquirer: Selection Bias
Acquirer: Conflict of
Interest Target: Before Positive Positive Negative Negative Target: After Positive Positive Negative Negative Acquirer: Before Negative Negative No Bias Positive Acquirer: After Negative Negative No Bias Positive Predictions of Analyst Bias Under the Selection Bias and Conflict Of Interest Hypotheses in Cash Deals. Predictions are sorted by analyst affiliation, whether the reports are on the acquirer or target, and whether the report is issued before or after the transaction.
42
Table 2: Statistics On Deals
Deals Where Estimates are Available for Target or Acquirer Currency # Deals Mean
Value Stdev 25% Median 75% Min Max
Cash 900 635 1,417 72 198 613 1 25,065 Stock 1,655 1,617 6,886 64 202 790 2 164,746
Deals Where Estimates are Available for Acquirer Currency # Deals Mean
Value Stdev 25% Median 75% Min Max
Cash 649 559 1,431 53 164 474 1 25,065 Stock 1,606 1,593 6,934 64 201 768 2 164,746
Deals Where Estimates are Available for Target Currency # Deals Mean
Value Stdev 25% Median 75% Min Max
Cash 586 888 1,681 155 360 938 12 25,065 Stock 901 2,689 8,842 214 561 1,832 6 164,746
Summary statistics on the aggregate transaction values of all the deals in our database in where information on fees as well as EPS or growth forecasts were available for either the target, acquirer, or the parent s of the target or acquirer. Currency for a deal is classified as “stock” if 50% or more of the consideration was paid for in stock and “cash” otherwise.
43
Statistics on affiliated analysts’ deviation from consensus 1-Year ahead EPS forecasts. Data are segregated by affiliation, whose EPS is being forecast, and whether forecast is issued before or after deal. A deal is a stock deal if 50% or more of the deal value is paid in stock; otherwise it is a cash deal.
Panel A: Cash Deals, 1 FY ahead forecast
Analyst Affiliation
Subject of
Forecast
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 496 0 0 -0.04% 0.22 -0.16% 0.00% 0.10% -5.23% 3.70%Target Acquirer After 506 0 0 -0.03% 0.49 -0.20% 0.00% 0.16% -4.55% 15.33%Target Target Before 425 + + -0.05% 0.61 -0.21% 0.00% 0.14% -18.46% 30.21%Target Target After 15 + + 0.03% 0.76 -0.24% 0.11% 0.25% -0.64% 0.62%
Acquirer Acquirer Before 598 0 + -0.07% 0.47 -0.23% -0.01% 0.15% -39.08% 21.57%Acquirer Acquirer After 679 0 + -0.20% 0.01 -0.29% -0.04% 0.11% -43.39% 3.81%Acquirer Target Before 207 - - -1.94% 0.25 -0.21% -0.02% 0.13% -339.61% 46.47%Acquirer Target After 9 - - 0.44% 0.06 0.00% 0.41% 0.45% -0.22% 1.71%
Panel B: Stock Deals, 1 FY ahead forecast
Analyst Affiliation
Subject of
Forecast
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 1034 - ? -0.11% 0.03 -0.09% 0.00% 0.07% -48.74% 3.08%Target Acquirer After 1115 - + -0.03% 0.24 -0.14% 0.00% 0.09% -10.25% 14.46%Target Target Before 726 + + 0.10% 0.51 -0.19% 0.00% 0.15% -25.42% 84.75%Target Target After 87 + + 0.08% 0.63 -0.22% 0.00% 0.23% -4.40% 12.96%
Acquirer Acquirer Before 1129 + + -0.03% 0.14 -0.11% 0.00% 0.11% -5.42% 6.31%Acquirer Acquirer After 1205 + + -0.03% 0.25 -0.13% 0.00% 0.11% -14.03% 8.38%Acquirer Target Before 448 - - -0.07% 0.19 -0.19% -0.01% 0.11% -16.95% 7.05%Acquirer Target After 72 - - 0.00% 0.97 -0.10% 0.00% 0.10% -2.09% 3.83%
Table 3: Descriptive Statistics on Affiliated Analyst Relative Optimism/Pessimism
44
Panel C: Cash Deals, 2 FY ahead forecast
Analyst Affiliation
Subject of
Forecast
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 428 0 0 -0.03% 0.51 -0.30% 0.00% 0.25% -5.30% 6.67%Target Acquirer After 426 0 0 -0.13% 0.10 -0.30% -0.01% 0.24% -12.95% 7.27%Target Target Before 319 + + -0.09% 0.57 -0.32% 0.00% 0.35% -18.22% 31.43%Target Target After 12 + + 0.36% 0.40 -0.05% 0.22% 0.78% -2.67% 3.47%
Acquirer Acquirer Before 533 0 + -0.10% 0.49 -0.24% 0.00% 0.33% -55.00% 49.74%Acquirer Acquirer After 677 0 + -0.15% 0.10 -0.29% -0.06% 0.39% -47.99% 8.99%Acquirer Target Before 160 - - 1.06% 0.12 -0.27% 0.03% 0.36% -5.26% 104.15%Acquirer Target After 3 - - 0.21% 0.46 -0.25% 0.40% 0.47% -0.25% 0.47%
Panel D: Stock Deals, 2 FY ahead forecast
Analyst Affiliation
Subject of
Forecast
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 974 - - -0.03% 0.27 -0.16% 0.01% 0.17% -10.78% 8.28%Target Acquirer After 938 - + -0.01% 0.81 -0.19% 0.00% 0.20% -9.43% 15.37%Target Target Before 592 + + -0.15% 0.58 -0.24% 0.00% 0.26% -148.31% 40.68%Target Target After 60 + + 0.00% 0.99 -0.63% 0.08% 0.61% -6.19% 6.93%
Acquirer Acquirer Before 1039 + + -0.03% 0.40 -0.15% 0.01% 0.22% -13.02% 5.60%Acquirer Acquirer After 1040 + + 0.04% 0.25 -0.20% 0.00% 0.20% -8.56% 7.59%Acquirer Target Before 362 - - -0.37% 0.11 -0.25% -0.01% 0.17% -53.39% 11.86%Acquirer Target After 59 - - -0.04% 0.81 -0.53% -0.16% 0.19% -2.17% 4.96%
Table 3, Continued Descriptive Statistics on Affiliated Analyst Relative Optimism/Pessimism
Statistics on affiliated analysts’ deviation from consensus 2-Year ahead EPS forecasts. Deviation is defined as affiliated analyst forecast less consensus, scaled by the last available closing stock price. Data are segregated by affiliation, whose EPS is being forecast, and whether forecast is issued before or after deal. A deal is a stock deal if 50% or more of the deal value is paid in stock; otherwise it is a cash deal.
45
Panel E: Cash Deals, Long-Term Growth Forecast
Analyst Affiliation
Subject of
Forecast
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 51 0 0 -0.43 0.65 -3.20 -0.40 2.40 -22.00 21.00Target Acquirer After 47 0 0 1.22 0.19 -2.00 0.00 3.00 -20.00 20.00Target Target Before 28 + + 0.31 0.84 -4.50 -1.50 3.25 -15.00 25.00Target Target After 1 + + 10.00 . 10.00 10.00 10.00 10.00 10.00
Acquirer Acquirer Before 39 0 + 1.13 0.23 -1.50 0.25 4.00 -15.00 20.00Acquirer Acquirer After 45 0 + -1.20 0.18 -4.70 -1.00 3.00 -15.00 13.67Acquirer Target Before 17 - - -5.40 0.40 -3.80 0.50 2.00 -100.00 20.00Acquirer Target After 1 - - -26.00 . -26.00 -26.00 -26.00 -26.00 -26.00
Panel F: Stock Deals, Long-Term Growth Forecast
Analyst Affiliation
Subject of
Forecast
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 46 - ? 0.91 0.37 -2.00 0.00 2.50 -12.50 26.00Target Acquirer After 13 - + -1.66 0.36 -1.75 0.00 1.00 -20.00 5.00Target Target Before 81 + + -0.74 0.25 -2.50 0.00 2.00 -22.00 12.00Target Target After 136 + + 1.28 0.04 -2.00 0.00 3.38 -20.00 40.00
Acquirer Acquirer Before 29 + + 0.22 0.72 -1.00 0.00 3.00 -10.00 5.00Acquirer Acquirer After 7 + + 2.96 0.24 -1.00 0.55 5.00 -2.00 15.20Acquirer Target Before 0 - - 0.00 0.00 0.00 0.00 0.00 0.00 0.00Acquirer Target After 0 - - 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Table 3, Continued Descriptive Statistics on Affiliated Analyst Relative Optimism/Pessimism
Statistics on affiliated analysts’ deviation from consensus long-term growth forecasts. Deviation is defined as affiliated analyst forecast less consensus, scaled by the last available closing stock price. Data are segregated by affiliation, whose growth is being forecast, and whether forecast is issued before or after deal. A deal is a stock deal if 50% or more of the deal value is paid in stock; otherwise it is a cash deal.
46
Panel G: Cash Deals, Recommendations
Analyst Affiliation
Stock being
Recom-mended
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 68 0 0 0.00 0.97 -0.29 0.00 0.83 -4.00 2.00Target Acquirer After 81 0 0 -0.01 0.91 -1.00 0.00 1.00 -2.00 2.00Target Target Before 54 + + 0.06 0.66 -1.00 0.00 1.00 -2.00 2.00Target Target After 5 + + 0.70 0.18 1.00 1.00 1.00 -1.00 1.50
Acquirer Acquirer Before 65 0 + 0.17 0.16 -0.33 0.00 1.00 -2.00 2.00Acquirer Acquirer After 143 0 + 0.23 0.00 0.00 0.00 1.00 -2.00 2.00Acquirer Target Before 27 - - 0.29 0.16 0.00 0.00 1.00 -1.00 3.00Acquirer Target After 3 - - 1.56 0.00 1.50 1.50 1.67 1.50 1.67
Panel H: Stock Deals, Recommendations
Analyst Affiliation
Stock being
Recom-mended
Before/After Deal N
Prediction under
selection bias
Prediction under
Conflict of Interest Mean pvalue
25 pctile Median
75th pctile min max
Target Acquirer Before 174 - - 0.04 0.59 -1.00 0.00 1.00 -2.00 4.00Target Acquirer After 232 - + 0.12 0.09 -0.82 0.00 1.00 -2.00 3.00Target Target Before 101 + + 0.29 0.00 0.00 0.00 1.00 -3.00 2.00Target Target After 27 + + 0.15 0.43 -0.50 0.00 1.00 -1.50 2.25
Acquirer Acquirer Before 154 + + 0.18 0.02 0.00 0.00 1.00 -3.00 3.00Acquirer Acquirer After 241 + + 0.13 0.04 -0.50 0.00 1.00 -3.00 3.00Acquirer Target Before 56 - - 0.26 0.06 -0.50 0.00 1.00 -2.00 3.00Acquirer Target After 17 - - 0.74 0.00 0.00 1.00 1.50 -1.00 2.00
Table 3, Continued Descriptive Statistics on Affiliated Analyst Relative Optimism/Pessimism
Statistics on affiliated analysts’ deviation from consensus recommendation. Deviation is defined as affiliated analyst forecast less consensus. Data are segregated by affiliation, whose stock is being recommended, and whether the recommendation is issued before or after deal. A deal is a stock deal if 50% or more of the deal value is paid in stock; otherwise it is a cash deal.
47
Prediction 1-Year Ahead EPS 2-Year Ahead EPS Long-Term GrowthModel 1 Model 2 Model 1 Model 2 Model 1 Model 2
AFa + -0.0003 -0.0038 0.011 0.0037 -0.1164 -0.4255Standard Error 0.0006 0.0078 0.0243 0.0063 0.663 1.1081Pvalue 0.71 0.69 0.33 0.28 0.57 0.65
(1-A)Ft + 0.0002 0.0002 -0.0015 -0.0015 1.0052 1.005Standard Error 0.0006 0.0006 0.0025 0.0025 1.6329 1.6327Pvalue 0.35 0.35 0.72 0.72 0.27 0.27
AFt 0 0.0001 0.0001 0 -0.0002 -0.0975 -28.87Standard Error 0.0001 0.0001 0.0002 0.0002 0.2871 24.072Pvalue 0.36 0.71 0.92 0.53 0.74 0.24
(1-A)Fa - -0.0002 -0.0002 0.0024 0.0024 -0.015 -0.015Standard Error 0.0006 0.0006 0.0021 0.0021 0.1449 0.1449Pvalue 0.38 0.38 0.87 0.87 0.46 0.46
Itb - . 0.0016 . 0.0001 . 38.604Standard Error . 0.0033 . 0.0002 . 29.3Pvalue . 0.69 . 0.64 . 0.90
Ita + . 0.0003 . 0.0002 . 38.465Standard Error . 0.0002 . 0.0004 . 29.948Pvalue . 0.12 . 0.33 . 0.10
Iac + . 0.0022 . -0.0039 . -0.5814Standard Error . 0.0056 . 0.006 . 1.6956Pvalue . 0.35 . 0.74 . 0.63
Observations 164,928 164,928 143,354 143,354 12,579 12,579
Table 4: OLS Results
The above table presents results from OLS regressions in which analyst optimism in forecasts, in both EPS and long-term growth, are the dependent variables, and various interaction terms are the dependent variables. Control variables are also included. Fa are fees received from the acquirer. Ft are fees received from the target. A is a dummy equal to 1 if the acquirer’s growth or EPS is being forecast, and 0 otherwise. Itb = BASFt and Ita
= (1-B)ASFt, where S is the percent of deal value paid in stock and B is a dummy indicating whether the forecast was issued before the deal. Finally, Iac = A(1-S)Fa.
48
Prediction Model 1 Model 2 Model 3#strongbuy >1 1.207 1.207 1.207Standard Error 0.017 0.017 0.017Pvalue 0.00 0.00 0.00
#buy >1 1.068 1.072 1.072Standard Error 0.015 0.015 0.015Pvalue 0.00 0.00 0.00
#hold 0 0.761 0.763 0.763Standard Error 0.009 0.009 0.009Pvalue 0.00 0.00 0.00
#sell <1 0.548 0.549 0.549Standard Error 0.031 0.031 0.031Pvalue 0.00 0.00 0.00
#strongsell <1 0.645 0.653 0.653Standard Error 0.043 0.043 0.043Pvalue 0.00 0.00 0.00
AFa >1 . 0.980 0.983Standard Error . 0.020 0.022Pvalue . 0.84 0.78
(1-A)Ft >1 . 1.028 1.028Standard Error . 0.022 0.022Pvalue . 0.10 0.10
AFt None . 1.003 0.992Standard Error . 0.010 0.022Pvalue . 0.78 0.74
(1-A)Fa <1 . 0.904 0.904Standard Error . 0.041 0.041Pvalue . 0.01 0.01Itb <1 . . 1.006Standard Error . . 0.026Pvalue . . 0.59Ita >1 . . 1.019Standard Error . . 0.026Pvalue . . 0.24Iac >1 . . 0.989Standard Error . . 0.043Pvalue . . 0.60
R-square 3.893% 4.241% 4.245%Observations 26,863 26,786 26,786
The above table presents results from logit regressions in which analyst recommendations are the dependent variables, and various interaction terms are the independent variables, as well as the frequency of unaffiliated colleague recommendations of various types (#strongbuy, #buy, #hold, #sell, and #strongsell). Control variables are also included. Fa are fees received from the acquirer. Ft are fees received from the target. A is a dummy equal to 1 if the acquirer’s stock is being recommended, and 0 otherwise. Itb = BASFt and Ita
= (1-B)ASFt, where S is the percent of deal value paid in stock and B is a dummy indicating whether the forecast was issued before the deal. Finally, Iac = A(1-S)Fa.
Table 5: Logistic Regression Results
top related